Hi Matthieu, MPI does not provide any guarantee on the completion order of the reception requests in this case. More generally, regarding waitAny, the MPI-3.0 specification says that "If more than one operation is enabled and can terminate, one is arbitrarily chosen". Thomas On 09/19/2013 04:31 PM, Matthieu Dorier wrote:

Hi,

I would like a precision on the semantics of MPI_Waitany in this situation:

- Process 1 does 2 MPI_Irecv with different tags A and B (the order doesn't matter); it then waits on the two requests with MPI_Waitany. - In the mean time, process 2 does an MPI_Send with tag A, then MPI_Send with tag B. Can I ensure that MPI_Waitany will always give me the index of the message corresponding to the first MPI_Send, i.e. the one with the tag A, even in situations where process 1 called MPI_Irecv on a tag B first?

Thanks

Matthieu Dorier PhD student at ENS Cachan Brittany and IRISA http://people.irisa.fr/Matthieu.Dorier

_______________________________________________ discuss mailing list [email protected] To manage subscription options or unsubscribe: https://lists.mpich.org/mailman/listinfo/discuss